In #20829, we fixed compositor recycling when switching between
BrowserViews, but it turns out that there is one additional case that we
need to handle. When we create a completely new BrowserView instance, it
starts of as visible (even when it hasn't been added to the window),
which means that it will need its own compositor instead of using the
recycled compositor.
To fix this, lets make BrowserViews hidden by default until they're
added to the window. See also #19988. This is a potentially breaking
change given that the initial value of `document.visibilityState` will
now be `hidden`, but given the experimental status of BrowserViews, I
think this is a fine change to make. The old behavior can be restored
with `webPreferences: { show: true }`.
Notes: Fix compositor recycling when creating new BrowserView
* fix(urlrequest): allow non-2xx repsponse results
- closes#21046
* test(net): add test cases to verify non-2xx body
* test(session): update spec to match clientrequest behavior
* test(net): update test cases to match clientrequest behavior
* spec: clean up async net spec
* Revert "fix: handle WM_GETMINMAXINFO instead of letting chromium do it (#19928)"
This reverts commit 27ce6a9cd3.
* fix: don't reset the width and height when correcting window placement
* fix: always use extraHeaders mode
* fix: clear pending callbacks
* fix: do not use "extraHeaders" for net module
* test: webRequest should be able to modify CROS headers
* chore: CROS => CORS
Co-Authored-By: Milan Burda <milan.burda@gmail.com>
* chore: CROS => CORS
Co-Authored-By: Milan Burda <milan.burda@gmail.com>
* Add GetApplicationNameForProtocol.
* Fix Windows implementation.
* Fix up test.
* Add documentation.
* Implement for real on Linux using xdg-mime.
Also ensure we allow blocking calls here to avoid errant DCHECKing.
* Improve docs for Linux.
* Clean up tests.
* Add a note about not relying on the precise format.
* Update docs/api/app.md
Co-Authored-By: Shelley Vohr <codebytere@github.com>
* Remove needless `done()`s from tests.
* Use vector list initialization.
* Add a simple test for isDefaultProtocolClient.
* Remove unneeded include and skip a test on Linux CI.
* We no longer differentiate between CI and non-CI test runs.
This was a regression in #16125, which unintentionally put
`GlobalShortcutListener::RegisterAccelerator` into a
`#if defined(OS_MACOSX)` block.
Notes: Fix broken `globalShortcut.registerAll()` on Windows and Linux
* fix: explicitly resize the contents when exiting html fullscreen while in OS fullscreen
* test: ensure HTML fullscreen toggles while in OS fullscreen
* refactor: use gin in Promise
* refactor: separate Promise impl that returns nothing
* refactor: use Promise<void> for promise that returns nothing
* fix: methods should be able to run on both browser and renderer process
* fix: should not pass base::StringPiece across threads
* refactor: no more need to use different ResolvePromise for empty Promise
* refactor: move Promise to gin_helper
* chore: add code required to use chromes spellchecker
* chore: fix linting
* chore: manifests needs buildflags now
* chore: add dictionarySuggestions to the context menu event when the spellchecker is active
* chore: enable by default for windows builds
* chore: add patch to remove incognito usage in the spellchecker
* chore: add dependencies on spellcheck common and flags
* chore: conditionally include spell check panel impl
* chore: fix deps for spellcheck feature flags
* chore: add patch for electron resources
* chore: add dependency on //components/language/core/browser
* chore: patches to make hunspell work on windows
* build: collect hunspell dictionaries into a zip file and publish
* chore: clean up patches
* chore: add docs and set spell checker url method
* chore: fix error handling
* chore: fix hash logic
* build: update hunspell filename generator
* fix: default spellchecker list to the current system locale if we can
* docs: document the language getter
* chore: patch IDS_ resources for linux builds
* feat: add spellcheck webpref flag to disable the builtin spellchecker
* chore: fix docs typo
* chore: clean up spellchecker impl as per feedback
* remove unneeded deps
* refactor: move the arg instead of const reference it
* refactor: avoid unnecessary copies of base::Value in arg
* refactor: pass-by-value in dict_util
* refactor: avoid unnecessary reference
* refactor: convert Menu and globalShortcut to gin
* refactor: convert api::Cookies to gin
* refactor: convert View and WebContentsView to gin
* refactor: convert WebContents related classes to gin
* refactor: convert powerMonitor to gin
* refactor: prepare for header change
* refactor: remove last uses of mate::EventEmitter
* refactor: remove mate::EventEmitter
* refactor: move trackable_object to gin_helper
* fix: custom converter should not use Handle
* fix: no more need to check if icon is empty
It was a bug that the Handle<NativeImage> can be non-empty when the
image file does not exist. The bug was caused by the converter code
writing out the image even when the convertion fails.
The bug was work-arounded by adding an additional check, but since the
original bug had been fixed, the additional check is no longer needed.
* fix: should always set frameId even when callback is null
* fix: do not mix gin/mate handles for NativeImage
* fix: set proper constraints for windows with defined maxWidth
When BrowserWindow has set constraints for width (max or min) it
won't behave correctly during first attempt of resizing it. When
maxWidth is defined and maxWidth equals its width it will shrink
rapidly when user tries to expand its width. On the other hand
when minWidth is defined and minWidth equals its width it's
possible to decrease its width with a few pixels.
Notes: Fixed improper behaviour of window with width constraint set during resize.
* fix: prevent crash when WM_GETMINMAXINFO is called on initing window
* refactor: remove direct uses of event_emitter_deprecated.h
* refactor: remove event_emitter_deprecated.h in api::App
* refactor: use std::move to save a copy
* fix: windows and linux builds
* refactor: remove a few uses of native_mate/gfx_converter.h
* refactor: deprecate mate::EventEmitter
* refactor: add gin_helper::EventEmitter
* refactor: convert a few classes to use gin_helper::EventEmitter
* refactor: get rid of native_mate_converters/gfx_converter.h
* fix: follow native_mate on reporting errors
* fix: gin is weak at guessing parameter types
* fix: incorrect full class name
* fix: gin::Handle does not accept null
* feat: add a new contextBridge module
* chore: fix docs linting
* feat: add support for function arguments being proxied
* chore: ensure that contextBridge can only be used when contextIsolation is enabled
* docs: getReverseBinding can be null
* docs: fix broken links in md file
* feat: add support for promises in function parameters
* fix: linting failure for explicit constructor
* Update atom_api_context_bridge.cc
* chore: update docs and API design as per feedback
* refactor: remove reverse bindings and handle GC'able functions across the bridge
* chore: only expose debugGC in testing builds
* fix: do not proxy promises as objects
* spec: add complete spec coverage for contextBridge
* spec: add tests for null/undefined and the anti-overwrite logic
* chore: fix linting
* spec: add complex nested back-and-forth function calling
* fix: expose contextBridge in sandboxed renderers
* refactor: improve security of default_app using the new contextBridge module
* s/bindAPIInMainWorld/exposeInMainWorld
* chore: sorry for this commit, its a big one, I fixed like everything and refactored a lot
* chore: remove PassedValueCache as it is unused now
Values transferred from context A to context B are now cachde in the RenderFramePersistenceStore
* chore: move to anonymous namespace
* refactor: remove PassValueToOtherContextWithCache
* chore: remove commented unused code blocks
* chore: remove .only
* chore: remote commented code
* refactor: extract RenderFramePersistenceStore
* spec: ensure it works with numbered keys
* fix: handle number keys correctly
* fix: sort out the linter
* spec: update default_app asar spec for removed file
* refactor: change signatures to return v8 objects directly rather than the mate dictionary handle
* refactor: use the v8 serializer to support cloneable buffers and other object types
* chore: fix linting
* fix: handle hash collisions with a linked list in the map
* fix: enforce a recursion limit on the context bridge
* chore: fix linting
* chore: remove TODO
* chore: adapt for PR feedback
* chore: remove .only
* chore: clean up docs and clean up the proxy map when objects are released
* chore: ensure we cache object values that are cloned through the V8 serializer
* refactor: convert methods of AutoUpdater to gin
* refactor: converter in map_converter.h is no more needed
* refactor: use gin in crash_reporter
* refactor: remove native_mate_converters/map_converter.h
* refactor: implement gfx_converter with gin
* refactor: convert methods of NativeImage to gin
* refactor: add gin_helper::Arguments
* fix: use gin_helper::Arguments to parse multi-type parameters
* refactor: use gin converters in api::Protocol
* refactor: convert JS constructor impl to gin
* refactor: use InitWithArgs helper
* fix: gin_helper::Dictionary should behave the same with mate
* fix cpplint warnings
* refactor: no more need to patch gin/dictionary.h
* refactor: use v8 serialization for ipc
* cloning process.env doesn't work
* serialize host objects by enumerating key/values
* new serialization can handle NaN, Infinity, and undefined correctly
* can't allocate v8 objects during GC
* backport microtasks fix
* fix compile
* fix node_stream_loader reentrancy
* update subframe spec to expect undefined instead of null
* write undefined instead of crashing when serializing host objects
* fix webview spec
* fix download spec
* buffers are transformed into uint8arrays
* can't serialize promises
* fix chrome.i18n.getMessage
* fix devtools tests
* fix zoom test
* fix debug build
* fix lint
* update ipcRenderer tests
* fix printToPDF test
* update patch
* remove accidentally re-added remote-side spec
* wip
* don't attempt to serialize host objects
* jump through different hoops to set options.webContents sometimes
* whoops
* fix lint
* clean up error-handling logic
* fix memory leak
* fix lint
* convert host objects using old base::Value serialization
* fix lint more
* fall back to base::Value-based serialization
* remove commented-out code
* add docs to breaking-changes.md
* Update breaking-changes.md
* update ipcRenderer and WebContents docs
* lint
* use named values for format tag
* save a memcpy for ~30% speedup
* get rid of calls to ShallowClone
* extra debugging for paranoia
* d'oh, use the correct named tags
* apparently msstl doesn't like this DCHECK
* funny story about that DCHECK
* disable remote-related functions when enable_remote_module = false
* nits
* use EnableIf to disable remote methods in mojom
* fix include
* review comments
* deprecate native_mate/native_mate/object_template_builder.h
* add gin_helper/object_template_builder.h
* add patch to avoid ambiguous error
* remove usage of object_template_builder_deprecated.h in a few files
* add note we should remove gin_helper/object_template_builder.h in future
* fix: use coordinate offsets in ShowAutofillPopup
* fix: silence coord int->float narrowing conversion
Minor fix to silence clang-tidy warning about implicit range narrowing.
Not a huge deal but the revision is also easier to read.
clang-tidy: bugprone-narrowing-conversions
* fix: emit updated on NativeTheme on the UI thread to avoid DCHECK
* Update atom_api_native_theme.cc
* spec: wait a few ticks for async events to emit so that test events do not leak into each other
Exposing these in the renderer didn't make sense as they weren't backed
by the same instance / value store. This API should be browser only
especially now that we have nativeTheme.themeSource. Exposing in
//common was a mistake from the beginning.
* refactor: fix clang-tidy vector operation warnings
Fix vector population performance-inefficient-vector-operation warnings
generated by clang-tidy
* refactor: fix clang-tidy emplace_back warnings
In cases where a temporary is created to be passed
to push_back(), replace it with emplace_back().
Warning: modernize-use-emplace
* refactor: fix clang-tidy loop iteration warnings
When practical, use range-based for loops instead of C-style for loops.
clang-tiny check: modernize-loop-convert
* refactor: fix clang-tidy string initialize warning
Remove redundant empty string initialization.
clang-tidy check: readability-redundant-string-init
